Elevenlabs

Eleven Labs Audio Generation

Generate high-quality audio using the Eleven Labs API.

Setup

Requires ELEVEN_API_KEY environment variable and uv package manager.

Scripts

All scripts are UV-based and self-contained. Run directly:

1. Text-to-Speech Narration

Convert text/markdown to speech with female voices.

scripts/narrate.py "Hello world" -o hello.mp3
scripts/narrate.py -f article.md -o article.mp3
scripts/narrate.py "Welcome!" -v rachel -o welcome.mp3

Voices: sarah (default), rachel, dorothy

2. Sound Effects Generation

Create one-shot audio effects from descriptions.

scripts/sound-effect.py "short notification ding" -o ding.mp3 -d 1.5
scripts/sound-effect.py "whoosh swoosh sound" -o whoosh.mp3
scripts/sound-effect.py "rain on window" -d 10 --loop -o rain.mp3

3. Voice Cloning

Clone a voice from audio samples (min 3 recommended).

scripts/clone-voice.py "My Voice" sample1.mp3 sample2.mp3 sample3.mp3

4. Speech-to-Speech (Voice Conversion)

Transform audio to use a different voice.

scripts/voice-convert.py recording.mp3 -o sarah_version.mp3
scripts/voice-convert.py speech.wav -v rachel -o rachel_version.mp3
scripts/voice-convert.py noisy.mp3 -o clean.mp3 --remove-noise

5. Music Generation

Compose original music from prompts.

scripts/compose-music.py "upbeat electronic background music" -o bg.mp3
scripts/compose-music.py "calm piano melody" -d 60 -o piano.mp3

6. Audio Isolation

Remove background noise from recordings.

scripts/isolate-audio.py noisy_recording.mp3 -o clean.mp3

7. Speech-to-Text Transcription

Transcribe audio with timestamps.

scripts/transcribe.py recording.mp3
scripts/transcribe.py podcast.mp3 -o transcript.json
scripts/transcribe.py video.mp4 --format srt -o subtitles.srt

8. Voice Design

Generate new synthetic voices from descriptions.

scripts/design-voice.py "A warm, friendly female voice with British accent"
scripts/design-voice.py "Deep male narrator" -t "Welcome to the show" -o preview.mp3

Voice IDs

Female voices:

• EXAVITQu4vr4xnSDxMaL — Sarah (soft, warm)
• 21m00Tcm4TlvDq8ikWAM — Rachel (calm, clear)
• ThT5KcBeYPX3keUQqHPh — Dorothy (pleasant)

Models

• eleven_flash_v2_5 — Fastest, 32 languages, lowest cost
• eleven_multilingual_v2 — Best quality, 29 languages
• eleven_turbo_v2_5 — Balanced speed/quality

Output Formats

• mp3_44100_128 — Good quality (default)
• mp3_22050_32 — Smallest file size
• pcm_16000 — Raw PCM for processing
• ulaw_8000 — Twilio compatible